The AI Daily Brief
Posts
ChatGPT's First True Competitor? Anthropic Releases Claude 2

ChatGPT's First True Competitor? Anthropic Releases Claude 2

Plus OECD says 27% of jobs at high risk of automation from AI

Nathaniel Whittemore
July 12, 2023

The AI Breakdown First Five - Wednesday July 12, 2023

Today on the First Five:

5. Follow Up: Classified Senate Briefing
4. Shutterstock and OpenAI Partner While Google Gets Sued
3. AI Designs Proteins In Seconds
2. OECD Says 27% of Jobs at Risk
1. Anthropic Releases Claude 2

5. Follow Up: Classified Senate Briefing

Yesterday the White House held its first classified AI briefing for Senators. The post-meeting reporting suggests that 1) the briefing reinforced the urgency of the issue for Senators; 2) that there is a profound tension between needing to set guardrails and not wanting to over-regulate when global competitiveness in AI seems strategically important; 3) there is surprising bipartisanship at the moment; 4) China is the boogeyman.

Senators leave classified AI briefing confident but wary of ‘existential’ threat posed by China

Senators departed their first-ever classified briefing on AI with a mostly bipartisan confidence in the United States' place in the world regarding the rapidly advancing technology.

www.foxnews.com/politics/senators-classified-ai-briefing-confident-wary-china-threat

4. Shutterstock and OpenAI Partner While Google Gets Sued

Everywhere you look there are lawsuits around AI companies training their models on copyrighted materials. Google, for example, was just hit with a class action lawsuit. OpenAI’s partnership with Shutterstock suggests a different approach to training data. Yesterday they extended their deal, and if nothing else, Wall Street liked it.

Shutterstock announces an expanded, six-year partnership with OpenAI that grants access to its video, image, and music libraries for #AI training; SSTK jumps 9%+
— Brian Solis (@briansolis)
5:49 PM • Jul 11, 2023

3. AI Designs Proteins In Seconds

In a recent interview, Google DeepMind CEO Demis Hassabis argued that although more people knew applications like ChatGPT and Bard, in his estimation AIs for things like medical technology were even more societally impactful. Case in point: this AI called RFDiffusion which can design proteins in seconds, which could be transformative in areas such as drug discovery.

I mentioned over the weekend that we were on the very cusp of being able to use AI for protein folding? Well, that shoe dropped today:
— Candice H. Brown Elliott (@Display_Geek)
10:30 PM • Jul 11, 2023

2. OECD Says 27% of Jobs at Highest Risk from AI

The OECD recently released its 2023 Employment Outlook. They surveyed 5300 workers across 7 OECD countries, and the big banner headline was that they found more than a quarter of jobs at the highest risk of being automated away by AI. Perhaps more nerve-wracking — the study came out before the rise of generative AI!

More than a quarter of jobs in the OECD rely on skills that could be easily automated in the coming artificial intelligence revolution, and workers fear they could lose their jobs to AI, the OECD. reuters.com/technology/27-…
— John Bailey (@John_Bailey)
11:58 AM • Jul 12, 2023

1. Anthropic Releases Claude 2

Anthropic has been coming up from behind OpenAI in a big way this year. That was cemented yesterday when they released Claude 2, a model that compares favorably to GPT-4 in many tests. Importantly, Claude 2 has more recent training data (going up to the beginning of this year); a much larger context window of 100,000 tokens (which is the equivalent of about 75,000 words — or one Great Gatsby); and is 4-5x cheaper. Hello!

Claude-2, Anthropic's shot at GPT-4, has arrived. It's cheaper than GPT-4 and far stronger in reasoning & coding than its older self.
Things you should know:
▸ On standard exams, it's not quite at GPT-4 yet but catching up fast compared to v1.3. Winner in bracket:
GRE verbal:… httptwitter.com/i/web/status/1…p
— Jim Fan (@DrJimFan)
4:52 PM • Jul 11, 2023